A low-order nonlinear model of a stalled airfoil from data: Exploiting sparse regression with physical constraints